Nvidia's Les Karpas on Robotics' ChatGPT Moment
· anime
The Robotics Riddle: Why the Industry’s Waiting for Its ChatGPT Moment
Les Karpas, Nvidia’s Global Head of Physical AI, is set to speak at TechCrunch Disrupt 2026. He will attempt to solve a long-standing mystery that has puzzled robotics enthusiasts and investors: what’s holding back the industry from achieving its full potential? The answer lies in the lack of an internet-wide dataset for physical AI.
The world was forever changed when ChatGPT burst onto the scene, demonstrating the power of large language models. Robotics, despite years of development and adoption, has yet to experience a similar breakthrough moment. Nvidia’s commitment to the field is clear – CEO Jensen Huang’s keynotes showcase the company’s enthusiasm for robotics – but even Karpas’ session can’t mask the elephant in the room: the industry’s inability to replicate the scale and data that made language models possible.
Companies like Shield AI, Colossal Biosciences, FieldAI, and Foxglove are pushing the boundaries of what’s possible with simulation, synthetic data, and foundation models. However, bridging the digital and physical is a significant challenge. The dataset issue requires a substantial effort to overcome.
Karpas’ unique perspective on this issue is due in part to his cross-disciplinary background – from architect to manufacturing engineer to startup CEO and venture capitalist. His experience speaks to the skills that robotics demands: creativity, technical expertise, and business acumen.
Nvidia’s role as a leader in AI research and development makes Karpas’ insights particularly valuable. However, it’s also worth asking what kind of breakthroughs would be required to put robotics on par with language models. The answer may lie not just in technological advancements but also in a broader shift in how innovation is approached.
Getting the right people together can make all the difference, as Karpas will attest. But the industry needs more than just inspiration; it requires a fundamental transformation in how companies collaborate and develop solutions to shared problems. This means embracing the complexity of physical AI and acknowledging the challenges that lie ahead. It also means recognizing the potential for growth and innovation within the industry – and being willing to take calculated risks to unlock it.
Investors and robotics enthusiasts must be willing to confront these complexities head-on, rather than relying on hypothetical solutions or silver bullets. By doing so, they can begin to unlock the full potential of physical AI and bring about a breakthrough moment for the industry.
Reader Views
- TIThe Ink Desk · editorial
The elephant in the room is not just the dataset issue, but also the economic reality of creating and collecting physical AI data at scale. Companies can't simply generate synthetic datasets forever; they need real-world examples to train their models on. What Karpas fails to address is how to incentivize companies to share their data and collaborate on large-scale projects, rather than hoarding it for competitive advantage. Until that hurdle is cleared, robotics will remain stuck in the slow lane behind language models.
- KAKenji A. · longtime fan
While Les Karpas' talk is a step in the right direction, I worry that Nvidia's focus on scaling robotics might overlook a more pressing challenge: ensuring these advancements aren't concentrated among a handful of tech giants. We've seen this play out with language models - their potential has been largely harnessed by large corporations and research institutions. Can we truly say the same for robotics, or will these breakthroughs be confined to a select few? It's time to consider the broader implications of AI democratization before we rush into scaling its applications.
- MPMira P. · comics critic
While Les Karpas is right to highlight the dataset gap in robotics, let's not forget that even with access to vast amounts of data, replicating language model success might be more complicated than meets the eye. Robotics relies on a fundamentally different type of data – messy, noisy, and variable across environments – which can't be easily simulated or generated. Companies like Shield AI are pushing the boundaries of synthetic data creation, but we need to consider the limitations and potential risks of relying too heavily on artificial environments. A breakthrough moment in robotics will require more than just a dataset; it demands new methods for bridging the gap between digital and physical worlds.